Closed Bug 1384364 Opened 7 years ago Closed 7 years ago

Intermittent Windows jsREFTEST NoneREFTEST ERROR | Got suite_end message before suite_start. Logged with data: {"thread": "

Categories

(Core :: Layout, defect, P5)

Product:
Core
Component:
Layout
Type:
defect

Tracking

()

Status:
RESOLVED INCOMPLETE

People

(Reporter: intermittent-bug-filer, Unassigned)

Details

(Keywords: intermittent-failure)

Filed by: wkocher [at] mozilla.com https://treeherder.mozilla.org/logviewer.html#?job_id=117548049&repo=autoland https://queue.taskcluster.net/v1/task/fstlapJ2RESslHvp0kNbfQ/runs/0/artifacts/public/logs/live_backing.log https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/fstlapJ2RESslHvp0kNbfQ/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1 I've seen a couple failures like this... One thing wrong with them is that the formatting is bad ("None" should be between '|' and properly spaced.). There's also probably a better thing than "None" to use here, like an actual test failure.
I am seeing something very similar with some Stylo Windows debug runs I am working on trying to enable soon: https://treeherder.mozilla.org/logviewer.html#?job_id=117749870&repo=try&lineNumber=18270 It's pretty mysterious, since the `suite_start` action gets logged a giant JSON object at the top of the log (which doesn't happen in a "good" run, which instead only prints "REFTEST SUITE-START | Running X tests"). :Callek, do you have any ideas about this? :KWierso thought you might know something.
Flags: needinfo?(bugspam.Callek)
I guess part of what's hard for me to follow is there's the "raw test log", with a name like "jsreftest_raw.log" which is just a bunch of JSON objects. The raw test log in good case[1] has `suite_start` recorded by source "reftest" which seems to be the logger in reftest.jsm. The raw test log in the bad case[2] has `suite_start` recorded by source "reftest harness" which seems to be `runreftest.py`. The "live log" that Treeherder links to by default only shows the JSON blob for suite_start in the bad case. So, it feels like for some reason in the bad case, the "reftest" logger in the test framework (reftest.jsm) isn't processing the suite_start, and instead we fallback to the harness. [1]: https://public-artifacts.taskcluster.net/Uo4aoaoiS6abv52RNW4VkA/0/public/test_info/reftest-stylo_raw.log [2]: https://public-artifacts.taskcluster.net/UxsDH63AR36w_oOQPJ7_1g/0/public/test_info/reftest-stylo_raw.log
I am also seeing this issue on some Linux runs, so I guess it's not Windows specific.
I have no knowledge about how jsreftest runs, so not really sure on the idea of logs being intertwined, perhaps someone in #taskcluster and/or #ateam can help there? That said there is some investigation into potential issues running jsreftest on w10 (no bug was filed yet) but I'm seeing if things are any better on win8 while that gets looked at.
Flags: needinfo?(bugspam.Callek)
https://wiki.mozilla.org/Bugmasters#Intermittent_Test_Failure_Cleanup
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.